Convolutional Neural Network for Speaker Recognition Embedding with Biometric System

Growing popularity of biometrics has led to concerns regarding the misuse and privacy of biometric data stored centrally. As a result of noise and variation between classes, biometric verification systems face challenges. Using an inter-modal biometric verification system, fingerprints and voice modalities can both be used. The System uses multiple biometric templates to combine the two modalities. Using finger print and voice data together effectively helps to minimize privacy concerns since the minutiae of a finger print are hidden among the artificial features derived from the speaker’s spoken utterance. This research work utilize Gaussian mixture models in conjunction with Universal Background Models (UBM). A pipeline for improving speaker recognition and Convolutional Neural Networks to learn speaker characteristics are also included. When compared with previous results for voice verification, the system database was used the accuracy of biometric identification is increasing overtime. On the whole, biometric identification can provide a secure identity management system for the global citizen that is fully independent. A majority of private employers also require biometric identification systems for access control to workplaces or time clocks that can't be tampered with. Student attendance is taken automatically through the proposed system.